A Comparison of Modeling Approaches for Stepped-Wedge Cluster Randomized Trials That Include Multilevel Clustering, Confounding by Time, and Effect Modification

• Activity Number: 163 - SPEED: Longitudinal/Correlated Data
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Lance Ford* and Julie A Stoner and Daniel Zhao and Tabitha Garwe and Ann Chou and Daniel Duffy
• Companies: University of Oklahoma Health Sciences Center and University of Oklahoma Health Sciences Center and OU Health Sciences Center and University of Oklahoma Health Sciences Center and University of Oklahoma Health Sciences Center and University of Oklahoma-Tulsa
• Keywords: Cluster randomized trials; stepped-wedge design; simulation studies; mixed effects models; longitudinal data analysis

Abstract

Stepped-wedge design (SWD), a type of cluster randomized trial, is being used more frequently to study community- and practice-based interventions. SWD is beneficial in cases where the intervention cannot be rolled out all at once and the intent is to expose all practices to an intervention program. Given the SWD is a relatively new design, optimal analytic approaches to account for correlation and time effects are not well established and have not been widely applied to data. This simulation study is designed based on a statewide intervention program called Healthy Hearts for Oklahoma. This study investigates how (1) levels of clustering and modeling approaches impact statistical efficiency of intervention effect estimates, (2) conclusions differ depending on the extent of extraneous temporal effects among modeling approaches, and (3) estimates vary across modeling approaches when modified by time under the intervention. Mixed effects models were used to account for the correlation between and within multiple levels of clustering. Programming simulations were conducted to estimate the bias and precision of estimates from different modeling approaches.